Texture Image Recognition Research Based On Bispectrum Slice

Texture recognition is regarded as an important branch of texture analysis, and the process plays an important role in many areas such as industrial automation, medical image analysis, document processing and bioinformatics. Texture is regarded as an important attribute and macroscopical representation of a certain local repetitive pattern based on the intensity of feature value in the image. However, due to the similarity and complexity of the repetitive pattern, it is difficult to describe the texture using languages. Texture image is different from the natural image, whose specificity is that each piece of texture image contains rich details and the same type of texture contains similar statistics characteristics. Therefore, it is more suitable for statistical recognition methods.At present, statistical recognition methods play an important role in texture analysis, in existing approaches, first, second order statistics have been widely used. But lower order statistics only contain the amplitude information of the texture image and neglect the importance of the phase information. However, phase of signal is very important, only using phase information of the image signal can be achieved for the complete reconstruction of the image. Therefore, when these methods are used to calssify texture images, which have the same second order statistics, they are usually inadequate to differentiate them. Compared with the first and second order statistics, higher order statistics can capture the spatial relationship between three or more pixels in the image. Most commonly used higher order statistics is the third order cumulants and bispectrum, because in practice there is all kinds of noise, in order to make the extracted features have certain anti-noise performance, bispectrum is selected to extract texture features.In this paper, taking it into account that the texture of different images may have the same second order statistics and limitations based on the recognition methods of traditional statistics, in order to get magnitude information and phase information of the texture image, we carry out an in-depth study analysis on higher order statistics. Therefore, the first step, bispectrum slices are used to extract texture features of the images, the second step is to use BP network based on the resilient BP algorithm as classifier. Finally, the simulation results demonstrate the feasibility and effectiveness of the experiment.